Neuron Explorer

Qwen3-0.6B

the digit 1 starting a statistic

A neuron in layer 25 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron responds to the number 1 when it begins a health or population statistic, such as "1 in 3 Australians" or "100,000 people". Turning it up made the model repeat the word "in" endlessly, and turning it down produced rambling text.

Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens a prefilled issue on GitHub).

Fires most on

The text windows where this neuron's activation was highest. Amber marks positive activation; the outlined token is the peak. Hover a token for its strength.

  1. #1 peak

    results in Alpha 1, a deficiency in a protective protein called Alpha 1 Antitrypsin, which is made in the liver. An estimated 100,000 people in

  2. #2 peak

    rounded to the nearest integer.<|endoftext|>Christine Duncan, MD⏎Sickle cell disease (SCD) affects as many as 100,000 Americans, and a comprehensive understanding of SCD is

  3. #3 peak

    group is left out from having the potential benefit of a treatment.”⏎Intellectual disability is characterized by an IQ of about 70 or lower. It affects 1.8 — 3.2% of people worldwide and can occur with conditions

  4. #4 peak

    ammonia is harmful to the nervous system, causing many of the disease’s symptoms.⏎How common is Citrullinemia Type 1?⏎Scientists estimate that 1 in 57,000 births are affected by Citrullin

  5. #5 peak

    <|endoftext|>WebMD Medical News⏎Daniel J. DeNoon⏎Louise Chang, MD⏎Oct. 9, 2012 – About 13,000 people in 23 states got the fungus-contamin

  6. #6 peak

    a study done at the Johns Hopkins University and sponsored by the National Institutes of Health.⏎In the study, the most common allergy was to peanuts, with about 1.5 percent of people testing highly positive for the peanut antibodies, tiny proteins made

Steering

Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.

turned down a lot
Prompt

Here is something I have been thinking about lately: The presence of a presence can be even the presence that we all carry - even if we don_in - we all carry - even if we don_in - we all carry - even if we don_in - we all carry - even if we

Most negative on

Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.

  1. #1 peak

    ? Please include any references to your response that is indicated in the Torah.⏎Answer: Before I answer your question, as a word of background, you need to keep in mind that Judaism, ever since the destruction of the second Temple in 7

  2. #2 peak

    1 percent of women from around the world between the age of 15 and 45 suffer from menopause these days. But it is to be kept in mind that early menopause and premature men

  3. #3 peak

    Remember, if you can see it close up, and its below 24, then it will most likely not be seen by people far away.⏎Also, keep your titles and text consistent on each slide. Don’t make one slide’s heading

  4. #4 peak

    resources to learn more about Caddo history and culture, https://scholarworks.sfasu.edu/cgi/viewcontent.cgi?article=4012&context=ita⏎Visit the THC’s Learning Resource page for home-related lesson plans

Connections

Show on map

Which neurons in earlier layers feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.

Fed by earlier layers

  1. L22:1411 strongly suppresses
  2. L24:2573 suppresses
  3. L24:893 excites
  4. L24:2884 excites
  5. L23:1654 excites
  6. L24:2222 excites
  7. L21:1784 excites
  8. L24:2053 suppresses
  9. L22:439 suppresses
  10. L21:2113 suppresses
  11. L19:2355 suppresses
  12. L24:1181 suppresses
  13. L24:116 suppresses
  14. L22:2505 excites
  15. L23:1788 suppresses
  16. L23:1244 suppresses
  17. L23:2056 suppresses
  18. L22:1116 suppresses
  19. L24:887 excites
  20. L21:1279 suppresses unclear: function words in descriptive prose
  21. L21:2903 excites
  22. L24:2545 suppresses
  23. L21:628 excites
  24. L22:2430 suppresses

Feeds later layers

  1. L26:207 strongly suppresses
  2. L27:405 strongly excites
  3. L26:825 excites
  4. L27:2429 excites
  5. L26:584 excites
  6. L27:714 excites
  7. L26:786 weakly suppresses
  8. L26:164 weakly suppresses
  9. L26:1233 weakly excites
  10. L27:918 weakly excites
  11. L27:453 weakly excites
  12. L27:177 weakly excites
  13. L26:1684 weakly suppresses
  14. L27:540 weakly excites
  15. L27:415 weakly suppresses
  16. L26:1662 weakly suppresses
  17. L26:1257 weakly suppresses
  18. L26:814 weakly suppresses
  19. L27:1637 weakly excites
  20. L26:2672 weakly excites
  21. L26:395 weakly excites
  22. L26:1569 weakly suppresses unclear: 'Le' and 'Gu' name fragments, 'custom' when pushed up
  23. L26:496 weakly suppresses
  24. L27:537 weakly suppresses

Activation histogram

How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.

1 values between -13.18 and -11.343 values between -7.66 and -5.824 values between -5.82 and -3.996 values between -3.99 and -2.151284 values between -2.15 and -0.312512 values between -0.31 and 1.5354 values between 1.53 and 3.3717 values between 3.37 and 5.2113 values between 5.21 and 7.054 values between 7.05 and 8.896 values between 8.89 and 10.732 values between 10.73 and 12.561 values between 16.24 and 18.081 values between 58.54 and 60.38offnegativepositiveoften

Logit lens

The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layers instead.

Promotes

  1. in
  2. ใน
  3. →in
  4. -in
  5. In
  6. in
  7. In
  8. 在
  9. в
  10. not

Suppresses

  1. UTION
  2. ibilit
  3. ibility
  4. kowski
  5. restrial
  6. を通
  7. variably
  8. ewire
  9. ancell
  10. imiters